Low-level Environmental Metals and Metalloids and Incident Pregnancy Loss

Abstract

1. Introduction

Contemporary human populations remain at risk for exposure to various environmental metals and metalloids through geographical [1], occupational [2, 3] or accidental contamination pathways [4]. Published reviews have focused on the reproductive and developmental toxicity (RADT) of individual metals and metalloids for human populations, including cadmium [5, 6], lead [7], mercury [8], and arsenic [9, 10]. Pregnancy loss is often considered when assessing the RADT of metal(loid)s, given that it represents embryonic or fetal mortality and has a high (25% to 31%) incidence when measured prospectively [11–14]. A recent review evaluated the collective evidence on the RADT of arsenic (As), cadmium (Cd), copper (Cu), lead (Pb), and mercury (Hg) and concluded that all but Hg were associated with spontaneous abortion [15]. Another review focusing specifically on low levels of metal(loid) exposures and male fertility concluded that evidence was strongest for Cd, Hg and Pb with less certainty for As [16]. Both reviews noted key methodologic limitations that impact the weighing of evidence, particularly very few prospective cohorts with exposures quantified during critical windows of human reproduction and development and adequate attention to potential confounders.

To date, we are unaware of research focusing on male and female partners’ exposures and pregnancy loss despite the couple dependent nature of pregnancy. In contrast to the ubiquitous nature of environmental metals and their designation as endocrine disrupters (EDCs) [17], or exogenous agents that can interfere with any aspect of hormone action [18], there are few epidemiologic studies focusing on metals at environmentally relevant concentrations. To date, much of our understanding about their RADT stems from an important literature focusing on higher exposed populations, such as occupational workers or residents of endemic geographic areas. Moreover pregnancy loss, per se, is not directly measured in most animal research, but rather relies on proxies such as resorption or late fetal deaths. Timing of demise is challenging to determine without prospective measurement. This data gaps precludes our understanding of the impact of metal(loids) on pregnancy loss among populations with environmentally relevant concentrations. Such concentrations are typical of those for most populations without unique occupational or geographical exposures, such as residence in an endemic region. In light of the importance of metal(loid) exposures during critical and sensitive windows of development for pregnancy outcomes, we sought to empirically assess these associations. Motivated by this gap, we sought to explore the individual relationships between environmentally relevant concentrations of three metals (cadmium, lead, mercury), one metalloid (arsenic) and incident pregnancy loss in a cohort of couples prospectively followed across sensitive windows of human reproduction.

2. Materials and Methods

2.5 Statistical analysis

We first characterized the study cohort by sociodemographic characteristics of partners with significance assessed using either the Chi-square or the Kruskal-Wallis tests. We estimated the distributions of all metal(oids) and summarized them as medians (Md) and interquartile ranges (IQR) with significance by partner estimated with the Wilcoxon-Mann-Whitney test. Time-to-pregnancy loss was estimated using Cox proportional hazard modeling techniques to estimate hazard ratios (HRs) and corresponding 95% confidence intervals for each metal(loid) and partner [28]. We defined time-to-loss as the number of days from observed ovulation as measured by the peak LH day detected by the fertility monitor to the date of reported loss. We used the peak LH day for estimated date of conception, given the short ( 24 hours) survival time of the ovum for fertilization. For 59 (17%) menstrual cycles without ovulation data in the pregnancy cycle, we used the day of LH peak from previous cycles as captured by the fertility monitor. For 16 women (5%) with insufficient ovulation data, we estimated conception as occurring 14 days prior to the first positive pregnancy test. The proportional hazards assumption was assessed using graphical checks based upon martingale residuals, and found to be met.

We modeled each partner’s metal(loid) concentration and time-to-pregnancy loss after log transforming (x+1) exposures to normalize distributions and rescaling by the standard deviation to aid interpretation of the hazard ratios (HRs) and 95% confidence intervals (CIs). Separate models were run for each metal(loid) and partner. Given the low correlation between couples’ metal(loid) exposures, we jointly modeled both partners’ metal(loid) concentrations relative to time-to-pregnancy loss. Specifically, Spearman correlation coefficients for couples’ exposures ranged from 0.26 to 0.28 and 0.40 for urinary Pb, Cd and As, and from 0.40 to 0.41 and 0.55 for blood Pb, Cd and Hg, respectively. We further adjusted all models for age (year, continuous), BMI (≤24.9 lean/normal, 25.0–29.9 overweight and ≥30.0 obese), average number of daily alcoholic beverages consumed and cigarettes (left continuous) smoked prior to conception for both partners and also during early pregnancy for female partners. Because of the uncertainty about the inclusion of prior loss in models, we ran models with and without prior pregnancy loss but conditioned on gravidity, as not all women had a prior pregnancy (no prior pregnancy, prior pregnancy but no loss, prior pregnancy with a loss). For the urine-based metal(loid)s, we also included the log-transformed creatinine concentration (mg/dL) as a covariate. The joint models included covariates for both partners with the exception of age, given their high correlation. Instead, we modeled one partner’s age and the difference in couples’ ages (in years). To ensure a complete analytic file and to minimize selection bias, we imputed missing metal(loid) concentrations and covariates using Markov Chain Monte Carlo methods under the missing at random assumption [29].

We assessed the linearity assumption underlying the association between metals and time-to-pregnancy loss using splines, and used inverse probability censoring weighting to correct for the couples who did not become pregnant. The weighting was performed in light of our earlier paper showing blood metals to be associated with a longer-time-to pregnancy [30], and to account for couples’ with higher exposures who did not become pregnant and, hence, are not included in the analysis. We used splines to assess any observed departures from linearity or thresholds and observed none (Figures 1 and ​and2).2). To this end, we used linear models for assessing exposures and pregnancy loss. We also conducted sensitivity analyses to assess the robustness of our findings. These included restricting our cadmium analysis to partners with concentrations >LOD, and also removing cigarette smoking as a covariate.

Open in a separate window

Figure 1

Shape of the associations between the female’s and male partners’ blood metal concentrations and hazard of pregnancy loss - splines.

Findings for female partners are presented in the left-sided figures, and those for males in the right-sided figures.

Open in a separate window

Figure 2

Shape of the associations between the female’s and male partners’ blood metal concentrations and hazard of pregnancy loss - splines.

Findings for female partners are presented in the left-sided figures, and those for males in the right-sided figures.

3. Results

Our cohort comprised mostly Non-Hispanic White couples who were highly educated, had household incomes ≥$70,000 and were covered by health insurance (Table 1). The mean ages of female and male partners were 29.8±3.9 and 31.6±4.6 years of age, and their mean BMIs suggested a heavier phenotype (27.0±6.7 and 29.7±5.5, respectively). The distributions for metal(loid)s by biospecimen are presented in Table 2 and reflect slight systematic differences in Md (IQR) concentrations by partner. Prevalence of cigarette smoking was low at enrollment, i.e., 7% and 11% of female and male partners, respectively. With the exception of blood cadmium, most partners had trace element concentrations above the LOD. The average time between enrollment and collection of biospecimens and pregnancy confirmation was 113 ± days.

None of the metal(loid)s irrespective of biospecimen (blood or urine) was associated with an increased risk of pregnancy loss even after covariate-adjustment (Table 3). In fact most adjusted HRs were around or below one, with only a few adjusted HRs being slightly above 1.0, i.e,, female partners’ blood Cd (1.08; CI 0.81, 1.44) and lead (1.01; CI 0.82, 1.25) concentrations. Only male partners’ blood Cd concentration unadjusted for covariates was positively associated with pregnancy loss (1.19; CI 1.01, 1.40), but the finding was not robust to covariate adjustment (1.09; CI 0.84, 1.41). When jointly modeling partners’ concentrations, females’ blood Cd (1.01; CI 0.75, 1.37) and both blood Pb (1.01; CI 0.80, 1.28) and urinary Pb (1.05; CI 0.83, 1.33) concentrations were positively associated with HR >1.0 when adjusting for covariates and the male partners’ concentrations. In joint models with both partners’ exposures modeled, none of the HRs were above 1 for male partners. Confidence intervals for all metal(loids) included one. Of note, the findings were robust in sensitivity analyses that did not include prior pregnancy history or without using inverse probability weighting (data not shown). The findings remained when removing smoking from the model or when restricting the Cd analysis to individual with concentrations >LOD (data not shown).

4. Discussion

To our knowledge, our findings are the first to describe the relation between environmental metal(loid)s in a cohort of couples without known unique geographical or occupational exposures, who were recruited prior to conception and followed daily through 7 post-conception weeks and monthly thereafter for the capture of incident pregnancy loss. We did not observe any significant associations between As, Cd, Hg, or Pb and pregnancy loss, and our findings were robust in sensitivity analyses. Other notable study strengths include having individually measured metals in blood and urine for both partners of the couple, along with our ability to detect losses over the course of gestation, including during the preconception and early pregnancy sensitive windows when most losses occur. Our overall incidence of loss was 28% with most losses occurring early and plateauing at approximately 72 post-conception days, as previously reported [11]. Our incidence rate is well within the range (25% to 31%) reported in earlier preconception cohort studies [12–14], increasing our confidence in having captured most losses after the post-implantation window. Still, we do recognize the potential for having missed preimplantation pregnancies or those losses that fail to meet hCG thresholds needed for home pregnancy tests. Nonetheless, our findings are strengthened by having prospectively measured pregnancy loss, rather than relying on retrospectively reported loss that has been shown to underestimate pregnancy loss and disproportionately early losses [31]. In light of 90% of women currently using home pregnancy tests for identifying pregnancy, we feel the findings have relevancy for contemporary populations [http://www.clearblueeasy.com/healthcare/heritage-of-clearblue.php] and are not limited by reliance on retrospectively reported losses. The lack of an association irrespective of measuring metals in blood or (along with As) in urine suggests that neither acute nor chronic exposures at these concentrations, respectively, are associated with loss. Also of note is the relatively comparable distribution of Pb, Cd, Hg and As exposures for our cohort in comparison with U.S. biomonitoring data [24].

Weighing our findings relative to the existing literature is challenging, given that most research has focused on subpopulations with much higher concentrations in light of unique environmental exposures, and most previous research has relied upon self-reported pregnancy loss that may not reflect the continuum of loss across gestation [31]. Our cohort’s exposures were skewed toward lower concentrations, which may explain the lack of observed associations with metal(loids). In the As endemic areas of Bangladesh and West Bengal, India, results from prospective studies of pregnant women did not reflect higher odds of spontaneous abortion even for women with the highest urinary (>200 ug/L) concentrations [32, 33]. However, the latter study did observe an association for stillbirth (OR=6.07; 95% CI 1.54, 24.0). In our study, all losses occurred by 22 weeks precluding our ability to assess stillbirths. No association was noted between As concentrations between 0.8 and 1.9 μg/L in drinking water in a U.S. study and spontaneous abortion [34]. Another study compared estimated As concentrations in drinking water for women with and without pregnancy loss at 5–20 weeks gestation and reported no association for concentrations ranging from 0.0 to 175.1 μg/L [35]. A recent meta analysis assessing As and spontaneous abortion reported greater odds for high (≥50 μg/L) and low-to-moderate (<50 μg/L) concentrations (OR 1.98; 95% CI 1.27, 3.10 and OR 2.02; 95% CI 1.40, 2.91, respectively) [10]. However, none of the odds ratios for the three studies with individually measured low-to-moderate doses achieved significance. Also of interest is the added attention the authors paid to methodology in this review. While the overall odds ratio achieved significance (OR 2.02; 95% CI 1.40, 2.91) even when restricting to studies with biomarker or individual data (2.20; 1.04, 3.46), significance was lost when relying on prospective cohort studies (1.45; 0.99, 2.12) or published research deemed of high quality (>7 on Newcastle-Ottawa Scale) (1.45; 0.99, 1.12).

While small, we observed the highest HRs for female blood Cd concentration and pregnancy loss. Of the few studies focusing on Cd and pregnancy loss, one relied on residence in a polluted area as a proxy of exposure while the other measured blood concentrations in 69 pregnant women. A higher prevalence of reported pregnancy loss was observed for women residing in the contaminated versus uncontaminated area (10% and 2.8%, respectively) [36]. In another study, Nigerian women reporting 3+ prior pregnancy losses had a higher mean blood cadmium level than women without a history (0.46 and 0.25 μg/L; p<0.05) [37]. These Cd concentrations were within range for women in our cohort.

Some of the early evidence in support of lead and pregnancy loss stems from a case control study of pregnant women whose blood lead levels were measured at varying times before 12 weeks gestation with telephone follow-up to ascertain loss [38]. The risk of spontaneous abortion was reported to double per 5 μg/dL increase in blood Pb (OR 1.8; 95% CI 1.1, 3.1). Careful interpretation of the findings is needed in light of a low prevalence of loss (5%). An older review of the literature focusing on Pb exposure and spontaneous abortion concluded that Pb exposure >30 μg/dL in men and between 5–9 μg/dL in women might be associated with an increased risk of spontaneous abortion [7]. In a cross-sectional study of pregnant women enrolled before 14 weeks gestation who were queried on prior history of spontaneous abortion, a 0.1% increase in the plasma/blood Pb ratio was associated with an increase (incidence rate ratio 1.18; p<0.05) in spontaneous abortion history [39]. However, no data were reported on measured plasma and blood Pb concentrations and outcome for the index pregnancy. A recent study focusing on low lead exposure in an occupational cohort of ceramic workers reported a positive trend between regular biomonitoring of exposure and miscarriage in the following 12 months, though prevalence was only 5% [3].

We were unable to locate data focusing on couples’ mercury exposures and pregnancy loss. An earlier and frequently cited occupational study reported that male partners’ urinary Hg concentration >50 μg/L was associated with a doubling (OR=2.26; 95% CI 0.99, 5.23) of the odds of spontaneous abortion in female partners [40]. In some studies, occupational exposure to metallic Hg among dental workers has been linked to menstrual disorders and spontaneous abortion when measuring exposure in hair [41].

Our findings do not support an association between either blood or urinary metal(loid)s at environmentally relevant concentrations and pregnancy loss, corroborating earlier studies reaching similar conclusions for lower relative to higher exposures that might be more reflective for general populations. These findings should not be interpreted as suggesting no potential RADT, as our earlier study focusing on three blood metals reported female cadmium and male lead levels at the environmental concentrations to be associated with diminished fecundity, as measured by a longer time to pregnancy [30]. However, collectively, our findings suggest that specific trace elements may diminish fecundity, but if couples are able to achieve pregnancy they are not at higher risk of loss. This is the rationale for our use of inverse probability weighting to account for couples not achieving pregnancy and who are systematically excluded from the analysis.

Cautious interpretation of our findings is needed in light of potential limitations including the possibility that some pregnancies and losses were missed, though we know of no a priori reason why such misclassification on outcome would be systematically associated with metal concentrations as couples did not know their levels. Model specification is always a concern but we observed little difference in point and interval estimates after adjustments, and our findings were robust in sensitivity analyses, including the removal of history of prior loss from models given the varying perspectives on how best to handle prior history. Another limitation is that we may be underpowered to detect small associations between the exposures and pregnancy loss in the context of our cohort size. Similarly, we may be underpowered to detect possible synergistic effects should they exist. Also, we only had measurements for AS and not metabolites. Perhaps future research capable of delineating mode of actions may help inform how best to weigh evidence from varying exposure profiles [42]. As reiterated for many environmental exposures, continued efforts to assess mixtures of EDCs including metals is needed keeping with the manner in which populations are exposed. The exposome conceptual framework is one such strategy and is well suited to the assessment of RADT [43]. A clearer delineation of the role of metal(loids) at low concentrations might aid communities’ concerns when accidental exposures arise, while working to develop public health initiatives consistent with calls about the preventability of miscarriage and stillbirth [44]. In conclusion, we found no evidence of an adverse relation between any of the metal(loids) considered in this work at environmentally relevant concentrations and incident hCG pregnancy loss.

Acknowledgments

Footnotes

References